Surprised it wasn't on the homepage yet, my X / LinkedIn feed is non stop about it.
I would love to hear from someone who actually played with all the FOMO harnesses / frameworks and actually did a first impression comparison on what to try out (I'm sure it's not even apples to apples) or discovered that it's all marginally better at best from just rolling your own via Claude Code / Codex / Cursor...
- paperclip
- hermes
- pi
- opencode
- openclaw
- nanoclaw
- gastown
- other FOMO framework I missed (not including skill frameworks such as gbrain etc)
OpenCode seemed perfectly workable as a programming assistant. As personal assistants, they all fall short. It's too difficult to really shape their output.
I was briefly impressed with OpenClaw a few times, but ultimately was turned off by not being able to get the models to stop being so damnably verbose. I thought I made progress for a while by having it tweak its soul, iterate, switch models, iterate, switch models, fuse the results, iterate... but ultimately it's all forgotten early in each session. And then one day it killed itself by rebuilding the container it was inside.
Hermes apparently has some plagiarism issues they're trying to cover up [0] and I was deeply unimpressed with their janky, flickery CLI that force-enables a bulky obnoxious header on every launch. Hermes did readily dive into its own source code and did readily confirm that there was no way to disable it. So that's neat. It constantly (wants to) run from upstream master which is unsettling.
Nanoclaw and nanobot seemed fine, but not notably different. There were some common bugs and glitches that caused some minor data loss while configuring nanobot. After that I just deciding to start hacking my own together.
What I really want in a harness is being able to actually control and rewrite the entire context window, like Zed's Text Threads before they obnoxiously and inexplicably removed what, to me, was their most powerful and distinguishing feature.
The timeline here is pretty telling and it looks like Hermes basically points their coding agent at evolver and says "reimplement this yourself." A few days later Hermes magically sports a nearly identical feature.
Check out Eve (https://eve.new). I saw it here on Hacker News after I decided OpenClaw was too raw for me. Eve is amazing. My theory as to why it's so good is that the founders have a strong background in design -- they met getting their masters in design at UC Berkeley. Their secret sauce is the harness they've constructed IMHO.
Here is what Zach Dive said in the original HN announcement--
Eve is an AI agent harness that runs in an isolated Linux sandbox (2 vCPUs, 4GB RAM, 10GB disk) with a real filesystem, headless Chromium, code execution, and connectors to 1000+ services.
You give it a task and it works in the background until it's done.
I built this because I wanted OpenClaw without the self-hosting, pointed at actual day-to-day work. I'm thinking less personal assistant and more helpful colleague.
The main interface is a web app where you can watch work happen in real time (agents spawning, files being written, use of the CLI). There's also an iMessage integration so you can fire a task asynchronously, put your phone down, and get a reply when it's finished.
Under the hood, there's an orchestrator (Claude Opus 4.6) that routes to the right domain-specific model for each subtask: browsing, coding, research, and media generation.
For complex tasks it spins up parallel sub-agents that coordinate through the shared filesystem. They have persistent memory across sessions so context compounds over time.
I've packaged it with a bunch of pre-installed skills so it can execute in a variety of job roles (sales, marketing, finance) at runtime.
This was not a pleasant experience to register with and evaluate. Nowhere is a Linux-deployable harness mentioned. It seems to be a Perplexity-like offering.
On top of that, users logging in with Google OIDC cannot delete their accounts. Eve asks for the login password to confirm account deletion. Broken.
I see your point. The others are self-hosted, and you bring your own key. Personally, I like the convenience Eve's online account provides. It's dead simple to be up and off to the races, getting useful work done with little effort. After struggling with OpenClaw, it has been a very pleasant experience. Several people in our company are using it now.
> I see your point. The others are self-hosted, and you bring your own key.
More importantly, they don't require an account with a company you don't know how long it'll exist.
It's less about "self-hosted vs not" and more about "If I start relying on this tool and the company eventually gets bought/sold/disappears/burns up, what can I do?"
On top of that, they're using wildly wrong TLD, it's not a generic fun TLD to do whatever with, it's specifically for doing a "New Action" in an existing platform. Gives me the impression they don't really know what they're doing.
> I would love to hear from someone who actually played with all the FOMO harnesses
I tried claude, opencode, pi, hermes, openclaw.
Specifically, I tried Claude with Sonnet/Opus and GLM-5.1, and OpenCode with Sonnet/Opus (briefly, since it's a violation of services) and much more with GLM-5.1.
I'll say: Claude is the best overall. OpenCode has the best UI. Pi has something going for it (I embed it in my agents on my Multica kanban), being that it's programmable and extensible by design, it's also a CLI.
Hermes: Sluggish, very slow to start, a lot is going on in the background, didn't like it. Seems over-engineered, didn't use it long enough to evaluate its memory functions. I'd rather have full session logs rather than these MEMORY.md summaries of what a session did.
OpenClaw: Amazing in its novelty, hellish in its implementation. I tried to make an OpenClaw on a VPS capable of editing its own Nix config, and it sucked. Tried a few variations like NanoClaw. Much less fluid; not the same, which is probably a good thing, but what OpenClaw tries to deliver is this autonomous agent with full ability to edit itself as crazy as that is. If it were just less sluggish and capable of more self-modification off the bat. I mostly blame JavaScript/NPM here.
I gave my brother an OpenClaw and he tried to make it do things on his behalf. His final feedback before abandoning it: It feels like I need to know programming. I ask it for the daily weather report.
I don't really see how Pi is a FOMO framework. It's basically the minimal surface area framework for you to build other things on. I'm building some agent powered applications and I would have no ability to do that reliably on Claude Code/codex/cursor because they lack the integration depth I need, that Pi provides.
Yeah, I used that term loosely, for me FOMO is everything I don't know. It's not a FOMO framework to you (but your comment definitely adds to my FOMO, in a totally positive way I mean...)
Hermes is more general purpose, like openclaw. Pi and opencode are more specific to coding.
Hermes accomodates other coding agents pretty well and has in fact bundled skills for claude code and codex (for spawning subagents, delegating, etc.). They are not exclusive
I haven't tried these 'agent fleet' frameworks in depth and am not sure yet they are not just a gimmick. Both openclaw and hermes handle multi-agennt orchestration, fwiw. Gas town looks like a silly way to burn all your tokens in a day. Paperclip is just buggy. I'm waiting for them to become more mature
I use Hermes at home. Swapped out Openclaw for this. It seems to work better with smaller contexts, chuncking it up in smaller pieces.
I don't code with it, I use Claude for that. But what I do do: it's a sysadmin for my homelab. It has a read-only mcp server to check the k8s status and has it's own ssh access to fix stuff after I approve it per session.
It's magical. Each morning I get a small update whether the backup ran, if pods are stuck or behaving weirdly etc... Since the entire homelab is GitOps I can always reverse a change made by the agent.
I am now adding Nextcloud and moving calendar offline, from Google to my own hw. I barely touch it anymore, manually. In stead I sent a quick voice note on Telegram and 30 sec later I get a screenshot 'proving' it worked.
> This is an independent information website about Hermes Agent, an open-source autonomous AI agent by Nous Research. We are not affiliated with Nous Research or any other organization referenced on this site.
Looks suspicious as hell, all those "We're just information pointing to official docs" websites look suspicious as hell.
It really isn't clear this website isn't run by Nous Research themselves, and it's pretty clear the website is intentionally trying to deceive people into thinking this is an official resource. In the footer for example:
> Contact: Hermes Agent is an open-source autonomous AI agent by Nous Research that grows smarter the longer it runs.
Not until you go into the Terms of Service is it clear it isn't related to Nous Research at all, and nothing makes it clear who/what is really running this website.
Really wild trying to defend slop websites like these.
I have Hermes and set it up so that it works on a self-hosted Matrix instance. It was pretty simple to set up after some back and forth with a more powerful model. It's primarily how I interact with Hermes. I don't really use a command line on my computer with it.
Just open up a Matrix Element chat window and go at it. I've been replacing more and more apps on my phone with this setup, and it's pretty streamlined. It's nice knowing that everything is local. The Matrix server is local, the local AI inference is local, etc.
I’m not sure, but I integrated Claude with Slack (with my own app; not the default Claude app).
It’s a game changer. Claude is literally like a co-worker now. When I ask it for work, it creates a dedicated Slack channel that runs that session Id. I can resume from Cli or import cli sessions into slack.
Telegram makes it very easy to create a bot. Other systems are either inconvenient or actively hostile to new bots, especially if you're not a well-known business.
Probably easiest way to get started today with something like that, agree. IRC is also very nice for it, and stupidly simple protocol even the dumbest LLM could interface with, trivial to host your own server too if you don't want to use existing ones with private channels.
I would love to hear from someone who actually played with all the FOMO harnesses / frameworks and actually did a first impression comparison on what to try out (I'm sure it's not even apples to apples) or discovered that it's all marginally better at best from just rolling your own via Claude Code / Codex / Cursor...
- paperclip
- hermes
- pi
- opencode
- openclaw
- nanoclaw
- gastown
- other FOMO framework I missed (not including skill frameworks such as gbrain etc)
reply